Mechanisms of MPEG Stream Synchronizatio n
`
`G J L,u, H K Pung and T S Chu a
`Department of Information Systems and Computer Scienc e
`National University of Singapor e
`Singapore 051 1
`
`Abstract
`
`Media synchronization is an important issue in developing multimedia applications . MPEG is an
`
`international standard for coding moving pictures and associated audio for multimedi a
`
`applications . Coded audio, video and other data streams are multiplexed into an MPEG stream .
`
`We introduce the syntax of the multiplexed MPEG stream and explain the mechanisms used t o
`
`maintain media synchronization in a hypothetical model, system target decoder, in which it i s
`
`assumed that data transfer and decoding are carried out instantaneously . Then we propos e
`extensions to these mechnisms to achieve MPEG stream synchronization in a practical decoder .
`
`le Introduction
`
`MPEG (Motion Picture Expert Group) is an international standard for coding of moving picture s
`
`and associated audio for digital storage media up to about 1 .5 Mbit/s, prepared by ISO/IEC JT C
`
`1/SC 29/WG11[l] . The draft standard [1] was published as ISO/IEC 11172 . When referring to
`
`this standard, MPEG and ISO 11172 are used interchangeably . The standard consists of two parts .
`
`Part 1 is divided into three sections . Section 1, Systems, specifies the system coding layer . I t
`
`defines a multiplexed structure for combining elementary streams, including coded audio, vide o
`
`and other data streams, and specifies means of representing the timing information needed t o
`replay synchronized sequences in real-time . Section 2, Video, specifies the coded representatio n
`of video data and the decoding process required to reconstruct pictures . Section 3, Audio ,
`
`specifies the coded representation of audio data . Part 2 , conformance testing, specifies th e
`
`procedures for determining the characteristics of coded bit streams and for testing complianc e
`
`with the requirements stated in Part 1 .
`
`MPEG Video is a very important aspect of the ISO 11172 standard and has been discussed b y
`Gall [2] . This paper looks into the MPEG Systems, particularly the aspects related to medi a
`synchronization . Since coded MPEG stream consists of continuous media streams, i .e. video and
`audio, it is necessary to synchronize these media for real-time playback . There are two aspects o f
`continuous media synchronization : intra-medium synchronization and inter-medi a
`
`ACM SIGCOMM
`
`-57 –
`
`Computer Communication Review
`
`1
`
`SAMSUNG-1019
`
`
`
`

`

`synchronization . While intra-medium synchronization ensures the continuity for smooth playbac k
`
`of each medium, intermedia synchronization ensures synchronization between associated media .
`
`Thus in this paper media synchronization is defined as occurence in which each medium is playe d
`
`out at its fixed rate determined by the type of medium and/or the application concerned and th e
`
`specified/required temporal relationships among the associated media are maintained .
`
`The ISO 11172 standard specifies syntax and semantics based on a conceptual decoder model ,
`
`system target decoder (STD) . This model assumes that the multiplexed MPEG stream is stored o n
`
`a constant latency digital storage medium, and data transfer and decoding within the decoder ar e
`
`instantaneous . ISO 11172 stream is designed such that the STD will be able to decode and displa y
`
`elementary streams synchronously .
`
`However, in a practical decoder, data transfer and decoding cannot take place instantaneously .
`
`Since ISO 11172 stream is specified based on STD model, it is the responsibility of decoder t o
`
`compensate for these data transfer and decoding delays to ensure synchronization .
`
`In ISO 11172, the storage medium has a broad meaning, including CD-ROM, magnetic harddisk ,
`
`digital audio tape and computer networks etc . These storage media am of indeterministic nature i n
`
`terms of delay and transmission rate instead of constant latency as assumed in STD model . In
`
`order to feed the ISO11172 decoder with a stream of constant latency, a medium specific decode r
`is required . An appropriate buffer must be used in the medium specific decoder to smooth out th e
`transmission jitter in order to provide data to the ISO 11172 decoder at the required rate .
`
`The rest of the paper is organized as follows . Section 2 discusses the MPEG Systems in general .
`Section 3 presents the principle and mechanisms used in the IS011172 Systems to support medi a
`synchronization in STD . Section 4 discusses compensation of decoding delays in practica l
`decoders required in order to maintain synchronization . Section 5 discusses buffer requirement t o
`
`overcome the bursty nature of digital storage medium .
`
`2. ISO 11172 Stream
`
`While MPEG Video and Audio specify the coding of each individual stream, MPEG System s
`specifies the syntax and semantics of information that is necessary in order to reproduce one o r
`more MPEG audio or video compressed data streams in a system . An ISO 11172 stream consists
`of one or more elementary streams multiplexed together . A elementary stream consists of
`a
`number of access units (AU) . In the case of compressed audio an access unit is defined as th e
`smallest part of the encoded bitstream which can be decoded by itself . In the case of compresse d
`
`video an access unit is the coded representation of a picture . A decoded audio access unit o r
`
`decoded picture is called a presentation unit (PU) . In a coded video stream, there are three type s
`
`of access units : I-pictures, P-pictures and B-pictures . I-pictures are coded without referring t o
`
`ACM SIGCOMM
`
`-58--
`
`Computer Communication Review
`
`2
`
`
`
`

`

`– Pack 1
`
`1
`
`pack
`header
`
`system
`header
`
`packet 1
`
`packet 2
`
`packet n
`
`pac k
`header
`
`packet
`
`I
`
`IS011172
`end cod e
`
`pack start SCR mux_rate
`code
`
`packe t
`start code
`
`stream id
`
`packe t
`length
`
`STD
`buffer size
`
`tim e
`stamps
`
`- nothing o r
`PTS or
`-PTS &DTS
`
`Fig .! ISOI 1172 Stream Syntax
`
`other pictures . P-pictures are coded using forward prediction and 13-pictures are coded using bot h
`forward and backward predictions . Due to the way video stream is coded, picture order in th e
`coded stream may differ from the display order . The decoder must carry out re-ordering if neede d
`
`[ 1 ,2] .
`
`An ISO 11172 stream is organized into two layers : the pack layer and the packet layer. The pac k
`layer is for system operations and the packet layer is for stream specific operations . Fig.1 shows
`the syntax of an ISO 11172 stream .
`
`An ISO 11172 stream consists of one or more packs . A pack commences with a pack header an d
`is followed by zero or more packets . The pack header begins with a 32-bit start-code . The pac k
`header is used to store system clock reference SCR and bitrate information, mux_rate . The SC R
`is a 33-bit number, indicating the intended time of arrival of the last byte of the SCR field at th e
`input of the system target decoder . Mux_rate is a positive integer specifying the rate at which th e
`system target decoder receives the ISO 11172 multiplexed stream during the pack in which it i s
`
`included . The value of mux_rate is measured in units of 50 bytes/second rounded upwards . The
`value zero is forbidden . The value encoded in mux_rate field may vary from pack to pack in a n
`ISO 11172 multiplexed stream. The mux_rate value together with SCR value defines the arriva l
`time of each byte at the input to the system target decoder .
`
`Data from elementary streams is stored in packets . A packet consists of a packet header followe d
`by packet data. The packet header begins with a 32-bit start code that also identifies the stream t o
`which the packet data belongs . The packet header defines the buffer size required at eac h
`elementary decoder for smooth decoding and playback of the elementary stream . The packet
`header may also contain decoding and/or presentation time-stamps (DTS and PTS) that refer t o
`the first access unit in the packet . The purposes of DTS and PTS are discussed in next section .
`
`ACM SIGCOMM
`
`-59--
`
`Computer Communication Review
`
`3
`
`
`
`
`
`
`

`

`The packet data contains a variable number of contiguous bytes from the same elementary stream .
`A data packet never contains data from more than one elementary stream and byte ordering i s
`preserved . Thus, after removing the packet headers, packet data from all packets with a commo n
`
`stream identifiers are concatenated to recover a single elementary stream . The multiplex of
`
`different elementary streams is constructed in such a way (in terms of packet size and the relativ e
`
`placement of packets from different streams) as to ensure that the specified STD buffers do no t
`
`overflow or underflow .
`
`The system header is a special packet that contains no elementary stream data . Instead it indicate s
`decoding requirements for each of the elementary streams . It indicates a number of limits tha t
`
`apply to the entire ISO 11172 stream, such as data rate, the number of audio and video streams ,
`
`and the STD buffer size limits for the individual elementary streams . A decoding system may us e
`
`these limits to establish its ability to play the stream . The system header also indicates whether th e
`
`stream is encoded for constant rate delivery to the STD. The system header must be the first pac k
`of the ISO 11172 stream . It may be repeated within the stream as often as necessary . Repeat of the
`system header will facilitate random access . Real-time encoding systems must calculate suitabl e
`limits for the values in the header before staffing to encode . Non-real-time encoders may make
`
`two passes over the data to find suitable values .
`
`Up to 32 ISO 11172 audio and 16 ISO 11172 video streams may be multiplexed simultaneously .
`Two private data streams of different types are provided . One type is completely private and th e
`other follows the same syntax as audio and video streams . It may contain stuffing bytes, a buffe r
`
`size field, and PTS and DTS fields . The use of these fields is not specified in ISO11172 .
`
`The system specification does not specify the architecture or implementation of encoder o r
`decoders . However, bitstream properties do impose functional and performance requirements o n
`encoders and decoders .
`
`The MPEG Systems coding specification provides data fields and semantic constraints on the dat a
`stream to support the necessary system functions . These include the synchronized presentation o f
`decoded information, the management of buffers for coded data, and random access . Rando m
`access is made possible by repeated appearance of the information needed to start decoding, suc h
`as SCR, PTS and system headers, and use of I-pictures (pictures coded without referring to othe r
`pictures). Other functions are all related to the smooth and synchrony playback of coded stream s
`and are discussed in the next section .
`
`3. Synchronization in the System Target Decode r
`
`An ISO 11172 stream is constructed such that elementary streams will be synchronously decode d
`and presented by the STD and elementary input buffers in STD will never overflow or underflow .
`This section explains the mechnisms used to achieve this .
`
`ACM SIGCOMM
`
`-60–
`
`Computer Communication Review
`
`4
`
`
`
`

`

`uncoded - - -
`Assemble
`video
`pictures ,
`sample STC
`for PTS
`
`I
`
`I
`
`uncode d
`audio
`
`Assemble
`audio frames ,
`sample STC
`fsr '
`
`Video out
`
`Compare
`PTS an d
`Display
`
`Audio out
`- -
`
`Compare
`PTS an d
`Display
`
`Encode
`
`Multiplexin g
`an d
`Buffer
`
`STC
`
`ISO 11172 Encoder
`--
`-—--—--—- -
`
`ISO 11 l' 2beco&r
`
`Decode
`
`Buffer
`
`Decod e
`
`Buffer
`
`S
`
`Fig.2 Protypical Encoder and Decode r
`
`ample STC
`fo r
`SCR
`
`Digital
`Storag e
`Mediu m
` L~ ?_ _
`
`Syste m
`decode and
`extract SCR
`
`In STD, playback of N streams is synchronized by adjusting the playback of all streams to a
`master time base rather than by adjusting the playback of one stream to match that of another . The
`master time base may be one of the N decoders ' clock, the DSM or the channel clock, or it ma y
`be some external clock . The similar synchronization principle has been used in a number o f
`synchronization schemes, such as synchronization tnarker[3], logical time system [4] and relativ e
`time system [5] . In these scheme, each presentation unit has a time stamp and presentation unit s
`with the same time stamps are displayed at the same time to achieve synchronization .
`
`MPEG Systems provide for end-to-end synchronization of complete encoding and decodin g
`process . This is achieved by use of time stamps, including system clock reference (SCR) ,
`presentation time stamp (PTS), decoding time stamp (DTS) . This end-to-end synchronization i s
`illustrated in Fig .2, which includes a protypical encoder (upper part) and a protypical decode r
`
`ACM SIGCOMM
`
`-61 —
`
`Computer Communication Review
`
`5
`
`
`
`
`
`
`
`
`

`

`(lower part) . These protypical encoding and decoding systems are not normative, they illustrat e
`
`the functions expected of real systems .
`
`In the protypical encoding system, there is a single system time clock (STC) which is available t o
`the audio and video encoders . Audio samples entering the audio encoder are organized into audi o
`present units . Some, but not necessary all, of the audio PUs have PTS values associated wit h
`
`them, witch are samples of the STC at the time the first sample of the PU is input to the encoder .
`
`Likewise, the STC values at the times when video pictures enter the video encoder are used t o
`
`create video PTS fields. SCR values represent the time when the last byte of the SCR field leave s
`
`the encoder . DTSs specify the time the access units are decoded, that is the time at which all th e
`
`bytes of an access unit are removed from the buffer of an elementary stream decoder in the ST D
`
`model . The STD model assumes instantaneous decoding of access units . In audio streams, and for
`
`B-pictures in video streams, the decoding time is the same as the presentation time and so only th e
`
`PTSs are encoded ; DTS values are implied . In video streams, for I-pictures and P-pictures th e
`
`DTS values are nominally equal to the PTS values minus the number of picture periods of vide o
`
`reordering delay multiplied by the picture period .
`
`In the protypical decoder system, the ISO 11172 stream arrives according to the arrival schedul e
`specified by SCR and mux_rate fields in the pack header . The first SCR value is extracted an d
`
`used to initialize the STC in the decoder . The correct timing of the STC is maintained by ensurin g
`
`that STC is equal to the subsequent SCR values at the time the SCRs are received . The STC ma y
`
`be maintained either by updating the STC with the value of the SCRs or via a control loop, usin g
`
`the SCR values as reference inputs . Elementary access units are decoded at times specified b y
`
`their DTSs and PUs are presented when their PT'S values are equal to the STC value . In this wa y
`
`both intra-stream and inter-stream synchronization is maintained . Intra-stream synchronization i s
`
`maintained by ensuring the STCs at the encoder and the decoder running at the same rate .
`
`Inter-stream synchronization is maintained by present each PU at their specified PTS relative t o
`STC. So long as each PU is presented at its specified time, the inter-stream temporal relationshi p
`is maintained .
`
`As mentioned in the last section, SCR is encoded in the pack headers . PTS and /or DTS are
`
`encoded in the packet headers . The PTS and for DTS are associated to the first access uni t
`
`commencing in the packet . More than one access units can commence in a packet . It is no t
`
`necessary for each packet to contain PTS and /or DTS . PTS and DTS fields are not necessaril y
`encoded for each picture or audio PU . They are requited to occur with intervals not exceeding 0 .7
`seconds . This bound allows the construction of a control loop using the PTS values which ha s
`
`guaranteed stability with a known bandwidth . For those PUs for which PTS is not encoded, th e
`
`decoder can approximate the correct value as the sum of the most recent PTS and an increment .
`
`The increment is the nominal number of system_clock_frequency cycles per PU times the numbe r
`
`of PUs since the last PTS .
`
`ACM SIGCOMM
`
`-62–
`
`Computer Communication Review
`
`6
`
`
`
`

`

`The value of the system clock frequency is measured in Hz and shall meet the followin g
`
`constraints :
`
`90 000 - 4 .5 <= system_clock_frequency <= 90 000 + 4 .5
`
`rate of change of system_clock_frequency with time <=250 * 10 -6 Hz/s
`
`All timing information (SCR, PPS, DTS) is specified in terms of a 90 kHz clock, which provide s
`
`sufficient accuracy for audio interchannel phase alignment . 90 kHz is a multiple of the variou s
`
`video frame rates under consideration and is also a submultiple of the CCIR 601 video samplin g
`
`frequency of 13 .5 MHz . The time stamps are encoded into 33 bits which are long enough t o
`
`support absolute program durations of at least 24 hours .
`
`So far, we have discussed about synchronization and time stamps . Buffer management is also a n
`important aspect of synchronization because if there is data starvation or buffer overflow ,
`synchronization will not be maintained . In ISO 11172, it is specified that for all multiplexe d
`
`streams the delay caused by system target decoder input buffering shall be less than or equal t o
`
`one second . The input buffering delay is the difference in time between a byte entering the inpu t
`
`buffer and when it is decoded . This STD buffering delay, together with elementary stream bitrate ,
`
`bounds the size of STD buffers . These buffer size values are specified in the system header an d
`packet header . The STD model precisely specifies the times at which each data byte enters an d
`leaves the buffer of each elementary stream decoder in terms of a common system time clock . It i s
`guaranteed that the input buffers with specified sizes will not overflow or underflow durin g
`decoding as long as the data stream conforms to the specification, and the complete decodin g
`
`system is synchronized in terms of SCR, DTS and PTS .
`
`To Summarize, MI'EG Systems specifies syntax and semantics of multiplexed stream based o n
`STD model such that STD can decode and present elementary streams synchronously .
`
`4. Compensation of Actual Decoding Delay s
`
`In the last section, we discussed synchronization provision for STD . But practical system does no t
`conform to STD model, because it takes time to transfer data from buffer and decode the data .
`This section looks at how to compensate for these delays .
`
`In a practical decoder, presentation time and decode time are not the same any more even fo r
`audio PUs and 8-pictures, because in practical decoding system decoding takes some time instea d
`of zero time as assumed in STD model . Furthermore, it takes different time to decode acces s
`units of different types of elementary streams . To compensate these differences, time stamps PT S
`and DTS have to be modified in a practical decoder . Assume them are two elementary streams ,
`one audio and one video, in an ISO 11172 stream . Let the decoding time for an audio access uni t
`
`ACM SIGCOMM
`
`-63–
`
`Computer Communication Review
`
`7
`
`
`
`

`

`be ta and decoding time for a video access unit tv . Also assume tv is greater or equal to ta . Then
`DTS and PTS should be adjusted in the decoder as follows :
`
`(1) In the ISO 11172 stream, DTSs are implied for audio access units and B-pictures . To mak e
`this explicit, a DTS is inserted immediately after each PTS with the same value . These DTS s
`indicate commencing times access units are being decoded . For those access units without DTS s
`
`associated, the decoding commencing time is equal to the most recent value of DTS plus tim e
`
`corresponding to access units between them .
`
`(2) To compensate for the decoding delay for video access units, the effective presentation tim e
`
`should be equal to the value of associated PTS plus decoding time tv .
`
`(3) For audio access units, if we choose the effective presentation time to be the value o f
`
`associated PTS plus decoding time ta, the audio presentation units will be presented tv-ta secon d
`
`ahead of video stream . This is not desirable . Therefore, in order to maintain the desired tempora l
`
`relationship between audio and video, the effective presentation time for audio access units shoul d
`be equal to the value of associated PTS plus tv . Decoded access units will be buffered at th e
`output buffer until their effective presentation times . The required output buffer is equal to th e
`
`audio display rate times (tv-ta) .
`
`Above discussion is concerned with decoding time and presentation time . Since it takes time to
`
`decode an access unit, the data corresponding to an access unit is not removed from th e
`
`elementary stream input buffer instantaneously . Therefore, in order to avoid buffer overflow, a n
`
`additional buffer size equivalent to the largest access unit in the elementary stream should b e
`
`added to the specified STD buffer size for each elementary stream .
`
`If presentation and decoding times and buffer sizes are adjusted as discussed above, a practica l
`decoder system will be synchronized and the buffer will not overflow or underfloor, provided tha t
`ISO 11172 stream data am retrieved from a constant latency storage medium . However, practica l
`digital storage media, such as harddisk and computer network, cannot deliver constant latenc y
`
`data stream . Therefore it is necessary to have a buffer in the storage medium specific decoder t o
`
`smooth the burstiness of the delivered data stream .
`
`5. Channel Smoothing and Storage Medium Requirement s
`
`An ISO stream can be coded in constant bitrate or variable bitrate . A bit, fixed_flag, in the syste m
`header indicates which mode is used . If its value is set to "1" fixed bitrate operation is indicated . If
`its value is set to "0" variable bitrate operation is indicated . To simplify the discussion in thi s
`section, a constant bitrate stream is assumed, although the principle also applies to the variabl e
`
`bitrate stream .
`
`ACM SIGCOMM
`
`-64 –
`
`Computer Communication Review
`
`8
`
`
`
`

`

`Under the above assumption, an ISO11172 decoder consumes data at a constant rate . However,
`digital storage media, such as harddisk and computer networks, are bursty in nature . Therefore, a
`buffer in the storage medium specific decoder has to be used to smooth the burstiness of dat a
`stream from the storage medium . The problem which we are interested in is to find what is th e
`
`buffer size requirement an d
`
`what requirements should be
`imposed on the storage Byte s
`medium so that the . storage
`
`A(t-tl )
`
`__ro
`
`medium specific decoder can
`
`output a constant latenc y
`
`constant rate data stream .
`
`Lets first examine requrement s
`
`on average data arriving rate a t
`
`the storage medium specifi c
`decoder and on the storag e
`medium transmissio n
`bandwidth . We introduce data
`arriving function A(t) and dat a
`
`output function C(t) . A(t )
`indicates the amount of dat a
`arrived at the storage mediu m
`specific decoder within tim e
`
`interval 0 to t . C(t) indicates the
`
`amount of data output from th e
`
`medium specific decoder
`within time interval 0 to t. Both
`A(t) and C(t) ar e
`non-decreasing function . C(t )
`increases with time at a fixe d
`output rate equal to the IS O
`11172 decoder consume rate .
`A(t) will normally not increas e
`
`at a fixed rate due to burstines s
`
`of the arriving rate caused b y
`delay jitter . Assume the firs t
`byte of data arrives at the clien t
`at time tl and the storage
`medium
`specific decoder
`
`t2
`ti
`Fig.3 A(t-tl) is close to C(t-t2), buffer requirement
`
`is small
`
`Byte s
`
`a
`tl
`Fig .4 Average arriving rate is larger than the output rate, buffe r
`requirement becomes larger and larger with tim e
`
`Bytes
`
`A(t-tl )
`
`tl
`
`2
`
`Fig.5 Average arriving rate is smaller than the output rate, there is a
`long initial delay and large buffer requiremen t
`
`ACM SIGCOMM
`
`-65–
`
`Computer Communication Review
`
`9
`
`
`
`
`
`
`
`

`

`outputs the first byte of data at time t2, then we have arriving function A(t-tl) and output functio n
`C(t-t2), as shown in Fig .3.
`
`In order to meet the continuity requirement, A(t-tl) must be equal to or greater than C(t-t2)
`
`. The
`
`difference, A(t-tl)-C(t-t2), is buffered and represents the buffer occupancy . The slope of A(t-tl )
`represents the data arriving rate . The average value of the arriving rate should be equal or close t o
`the output rate . If average arriving rate is greater than the output rate, the difference A(t-tl) an d
`
`C(t-t2), i .e . the buffer occupancy will become larger and larger (Fig .4) . This means that in orde r
`
`to successfully play out a stream, either a very large buffer size is required or playout of th e
`
`stream can only sustain for a short time . This is not desirable in a practical system .
`
`If the average arriving rate is smaller than the output rate, to satisfy the requiremen t
`A(t-t1)-C(t-t2)>= O, t2 must be large (Fig .5) . This means that the initial delay from the time th e
`first byte arrives to the time the first packet is displayed is very long . This also means that a very
`large initial buffer is required . The longer the stream to be played, the longer the initial delay an d
`the larger the buffer requirement . All these, long delay and large buffer size requirement, are no t
`desired and practical . Therefore the average arriving rate should be matched closely to the outpu t
`rate in order to output the data stream at constant rate continuously . To achieve this, the storag e
`transmission bandwidth at least equal to the consume rate must be guaranteed .
`
`Now lets look at the buffer size requirement in the storage medium specific decoder in order t o
`provide a constant latency stream to the ISO 11172 decoder . Suppose a byte can experience delay
`
`ranging from the minimum drain to the maximum d1
`from the data source to the storag e
`medium specific decoder . Then, if a byte experiences a delay of d is buffered for a time dmax d ,
`all bytes will experience the same total delay equal to dmax relative to the ISO 11172 decoder. I n
`this case, latency for each byte will be constant . The maximum required buffering time is dma x
`drain, which is the delay jitter. Therefore the buffer size requirement is equal to the arriving dat a
`
`rate times the channel delay jitter . It should be noted that the larger the delay jitter, the larger th e
`
`buffer size requirement . To limit the buffer size required at the storage medium specific decoder ,
`
`delay jitter should be small .
`
`Above discussion shows that a buffer size equal to bitrate times channel delay jitter is required t o
`
`smooth the channel and a channel bandwidth at least equal to the output rate must be guarantee d
`
`in order to play out a MPEG stream successfully .
`
`6e Summary
`
`MPEG Systems defines a multiplexed structure for combining coded audio, video and other dat a
`
`streams, and specifies means of representing the timing information needed to repla y
`
`synchronized sequences in real-time . The specification is based on a system target decoder model .
`
`ACM SIGCOMM
`
`-66–
`
`Computer Communication Review
`
`10
`
`
`
`

`

`We explained how media synchronization is achieved in the system target decoder . Then we
`discussed the required modification of some time stamps in a practical ISO 11172 decoder an d
`discussed channel smoothing buffer requirement in the storage medium specific decoder .
`
`References
`
`1. Draft International Standard ISO/lEC DIS 11172, 1992 .
`
`2. Didier Le Gall, "MPEG : A video Compression Standard for Multimedia Applications" ,
`
`Communications of the ACM, Vol .34, No .4, April 1991, pp .46-58 .
`
`3. D. Shepherd and M Salmony, "Extending OSI to support synchronization required b y
`multimedia applications", Computer Communications, Vol . 13, No . 7, September 1990 ,
`pp.399-406 .
`
`4. D. P . Anderson and G . Homsy, "A continuous Media I/O Server and Its Synchronizatio n
`
`Mechanism", Computer Journal, October 1991, pp .51-57 .
`
`5. P. V. Rangan, S . Ramanathan, H. M. Vin and T. Kaeppner, "Techniques for Multimedi a
`
`Synchronization in Network File Systems", Computer Communications, March 1993 .
`
`ACM SIGCOMM
`
`-67--
`
`Computer Communication Review
`
`11
`
`
`
`